Tokens, memory limits and long chats, in plain words
Tokens, context windows and limits, explained
AI chat tools count text in small chunks called tokens, and every model can hold only so many at once. These notes explain what that means, why a chat forgets and how usage limits and pricing work, using the makers' own documentation.
All the notes
- What is a token in an AI language model?
A token is the chunk of text a language model actually reads. Here's how tokens differ from words and why the difference matters.
- What is a context window?
A context window is the text a model can look at while it writes a reply. Here's what goes inside it and what sizes the makers publish.
- Why an AI chat forgets earlier messages
A chat assistant can lose earlier details when a long conversation fills its context window. Here's how that works, from the makers' own docs.
- How many words fit in a token budget?
A rule of thumb for turning tokens into English words and back, with a conversion table and the reasons it's only a rough guide.
- Input limits and output limits are different
Why a reply can stop short even when your document fit. The window, the output cap and hidden reasoning tokens are separate things.
- How token-based pricing works
How makers charge for AI use by the token, why input and output have different rates, and what cached input and reasoning tokens are.
- How to count tokens before you send
How to check the size of a prompt with the makers' own token counters, and why a plain-text count can miss part of a request.